(navigation image)
Home American Libraries | Canadian Libraries | Universal Library | Community Texts | Project Gutenberg | Biodiversity Heritage Library | Children's Library | Additional Collections
Search: Advanced Search
Anonymous User (login or join us)
Upload
See other formats

Full text of "Quantum Electronics-QBX-TMR Multi Function Timing Controller OCR"

Using Microprocessors to Learn About their Environment 



A.J. Kagan B.Sc. (Elec. Eng) M.B.A, 
Quantum Electronics 



One of the major advantages of microprocessors is that in any 
given project they allow a degree of "intelligence" and flex- 
ibilty not normally associated with conventional electronic con- 
trols. These functions may not be attainable at all with 
"discrete" electronics. Once the intelligence is present in the 
form of a microcomputer additional functions, outside those 
called for in the original specification, may be added at very 
little extra cost. In this paper I intend to present two examples 
of how this intelligence may be used to allow the microcomputer 
to learn about its environment. This results in one case as 
improved performance, and reduced maintenance, whilst in the 
other commisioning and duplication of the system is much simpli- 
fied. In order to place the area of discussion in perspective 
both projects as a whole are dealt with in some detail. 



Project 1 



Feeding a set amount of powder product (by weight) into a con- 
tainer poses several problems. The speed with which the powder is 
fed into the container directly affects the accuracy of the 
weight of the product which finally ends up in the container. 
This is a result of the amount of product in the air. Conven- 
tional processes (figure 1) weigh the product in a weigh pan and 
then release it down a chute into the container. Depending on the 
product this could stick in the chute. Further inaccuracies occur 
from the build up of product on the weigh pan. 

To solve these problems a different approach was suggested. The 
product (coffee) was to be fed into a rigid can. To gain speed 
the coffee was to be fed in at two separate stages. The first 
stage would allow a bulk fill to about 80% of weight and the 
second stage would trickle feed the remaining 20%. In between 
these two stages the product would be vibrated down to allow 
settling and make space for the additional 20%. The cans are 
queued and this allows the two processes to occur simultaneously. 
This solves the problem of inaccuracies due to the speed require- 
ments ("in flight" product). 

Being a rigid container each can is treated as a weigh pan. At 
the start of each feed cycle, each load cell (there are two 
required to attain the above process) is zeroed. This corrects 
for gradual build up of product of the load cell. The queue of 
cans is then shifted on. At the bulk fill stage the can is 
weighed so that at a later "stage" the amount of coffee in the 
can may be deduced. The product is then fed into the cans, which 
are resting on the load cells. As a result there is no product 
loss between the weigh pan and the can. 



1 



There are two stages of vibration between the bulk and trickle 
feeds, and the computer maintains a record of the can weights so 
that it knows this information when the can plus 80% of the total 
required weight arrives at the trickle feed stage. Figure 2 
depicts this process. 

To increase throughput 4 lanes run in parallel. The process runs 
under complete computer control, and the computer allows both 
random feeding of cans, and cans of varying weight within a 
particular range. It could also provide moving average informa- 
tion on the weights filled, and in the extreme even compensate 
the weights over a long term average to maintain the weights to 
the letter of the law, so there is an absolute minimum of give- 
away. This, however was not what I am trying to descibe, for the 
project was faced with a far more serious problem. 

The output of the load cell, in the millivolt range, was fed 
through amplifiers into a A to D converter. To say that these 
amplifiers were inadequate would be an understatement. Aside from 
the noise rejection, they would drift over a period of a few 
hours altering the gain and zero, after having been set to the 
same amplification. Of course there was also long term variation 
in the characteristics of the load cells which was to be expec- 
ted. The following solution could be applied, (somewhat less 
frequently) even in this case of only long term variation. 

This variation in the amplification is critical, not because of 
the variaion at a bulk or trickle stage individually, but because 
of the interaction between them. As mentioned above the weight of 
the can is measured on the bulk stage. This information is then 
used at the trickle stage. Originally this value was measured and 
used in the computer as volts. i.e. if the can generated 2 volts 
equivalent at the bulk stage it was considered to be 2 volts at 
the trickle. If the amplification had changed this value serves 
no useful purpose. 

The only way around this problem was by introducing a calibration 
mode where the computer would learn the characteristics of each 
load cell and its associated amplifer. Before going any further I 
would like to point out another advantage of using a micro- 
computer. No hardware change was required, (although was probably 
still desireable) all that was changed was the programme. 



The principle used is from high school mathematics. If one knows 
two points on a straight line graph then the equation of the line 
can be calculated. In the calibration mode two cans of known 
weight are fed through the machine. For each load cell the output 
voltages (in binary representation) are recorded, and then the 
constants in the equation of the form (y=mx+c) are calculated for 
all 8 load cells. Once the voltage of the can is read it can now 
be converted to actual weight. This actual weight is used for the 
trickle stage, bypassing the vagaries of variations in the con- 
version to a binary representation. The more frequently the 
calibration is done the more the production will maintain its 
accuracy. In fact the computer could go into the calibrate mode 



2 



Project 2 



periodically of its own accord. 



In railway 
post. This 
office to report 
information. Each 
normal telephone, 
with a particular 
calls iJn the 
panel with a 



telephone. The 
considerable 



S ince 

count 

puter 

and 

would 

Progr 

locat 

conne 

f utur 

would 

requi 

costs 



ry 



shunting yards the 
is to allow engine 
where they a 

telephone 
When the con 
phone this n 
relevant number is 
lamp at every loca 



system 
r of 



re is a 
driver 
re, pi 
has its 
trol ce 
umber i 
displa 
tion wh 
for a 



telephone 
s to call 
us to pass 

own numbe 
ntre wishe 
s "dialled 
yed. Ther 
ere there 
total of 5 



on each signal 
into the control 

on any relevant 
r, much like a 
s to communicate 
". When a driver 
e is also a mimic 
is a signal post 
12 telephones, a 



there are a number of shunting yards scattered around the 
it would be convenient if the central controlling com- 
could be adapted to allow different phone numbers to be set 
tq change location lamps on the mimic panel. Normally this 
be set up as a table of constants in EPROM (Eraseable 
ammeable Read Only Memory) . Each lamp would have a set 
ion to which it would have to be wired. If only 100 were 
cted, imagine the wiring problems that could arise. In 
e expansion each time a new telephone were to be added this 
require an EPROM change and replacement, and installation, 
ring switching off the system, new documentation, extra 
etc . 



Commisioning could also be a problem since the system has 9 

microcomputers and the setup according to a preordained (or 

possibly even unknown) set of telephone numbers could lead to 
severe problems. 

To provide a solution the central computer is made capable of 

learning. Each lamp is connected to any output just as long as 
the two wires are connected in the correct polarity. 

When a previously unrecognised signal post telephone calls in, 
the computer enters a special phase that allows the operator to 
enter the number he wants to be associated with that phone. The 
operatdr then steps through the mimic panel lamps until the 
required lamp is lit and instructs the computer that these are 
the parameters that are associated with the telephone. This 
information is then stored in some form of non-volatile memory. 

All location dependant parameters can be done on the spot, and by 
relatively unskilled operators, provided the "learn" routine is 
simple enough, e.g. operator prompting. 



3 



Summary 

In the first application I have tried to present how the computer 
can compensate for variation in its operating environment by 
using (certain stimuli and adjusting its operation to the respon- 
ses. The calibration procedure allows a non-skilled operator to 
calibrate the machine instead of a skilled technician to set up 
amplifications etc. 

In the second application I have tried to show that the computer 
can learn so as to simplify installation procedures, and to make 
a particular solution or product more adaptable to different 
locations. It also means that a less skilled person can configure 
a system. 

References : 

1 . Analog-to-digital converter calibration, A.J. Kagan, Pulse, 
July 1981. 

2. A computer controlled packaging machine, A.J. Kagan, Pulse, 
April 1982. 



For further information contact: 

Aubrey Kagan 
Quantum Electronics 
P.O. Box 391262 
Bramley 2018 



4 Highland 
173 

Orange 



Louis 



Phone (011) 7281241 



House (N.B.S. Centre) 
Botha Ave. 
Grove 



4